2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include "kerncompat.h"
21 #include <sys/ioctl.h>
22 #include <sys/mount.h>
25 #include <sys/types.h>
29 #include <uuid/uuid.h>
30 #include <linux/limits.h>
36 #include "transaction.h"
39 #include "task-utils.h"
42 #include <ext2fs/ext2_fs.h>
43 #include <ext2fs/ext2fs.h>
44 #include <ext2fs/ext2_ext_attr.h>
46 #define INO_OFFSET (BTRFS_FIRST_FREE_OBJECTID - EXT2_ROOT_INO)
49 * Compatibility code for e2fsprogs 1.41 which doesn't support RO compat flag
51 * Unlike normal RO compat flag, BIGALLOC affects how e2fsprogs check used
52 * space, and btrfs-convert heavily relies on it.
54 #ifdef HAVE_OLD_E2FSPROGS
55 #define EXT2FS_CLUSTER_RATIO(fs) (1)
56 #define EXT2_CLUSTERS_PER_GROUP(s) (EXT2_BLOCKS_PER_GROUP(s))
57 #define EXT2FS_B2C(fs, blk) (blk)
62 #define CONV_IMAGE_SUBVOL_OBJECTID BTRFS_FIRST_FREE_OBJECTID
65 uint32_t max_copy_inodes;
66 uint32_t cur_copy_inodes;
67 struct task_info *info;
70 static void *print_copied_inodes(void *p)
72 struct task_ctx *priv = p;
73 const char work_indicator[] = { '.', 'o', 'O', 'o' };
76 task_period_start(priv->info, 1000 /* 1s */);
79 printf("copy inodes [%c] [%10d/%10d]\r",
80 work_indicator[count % 4], priv->cur_copy_inodes,
81 priv->max_copy_inodes);
83 task_period_wait(priv->info);
89 static int after_copied_inodes(void *p)
97 struct btrfs_convert_context;
98 struct btrfs_convert_operations {
100 int (*open_fs)(struct btrfs_convert_context *cctx, const char *devname);
101 int (*read_used_space)(struct btrfs_convert_context *cctx);
102 int (*copy_inodes)(struct btrfs_convert_context *cctx,
103 struct btrfs_root *root, int datacsum,
104 int packing, int noxattr, struct task_ctx *p);
105 void (*close_fs)(struct btrfs_convert_context *cctx);
108 static void init_convert_context(struct btrfs_convert_context *cctx)
110 cache_tree_init(&cctx->used);
111 cache_tree_init(&cctx->data_chunks);
112 cache_tree_init(&cctx->free);
115 static void clean_convert_context(struct btrfs_convert_context *cctx)
117 free_extent_cache_tree(&cctx->used);
118 free_extent_cache_tree(&cctx->data_chunks);
119 free_extent_cache_tree(&cctx->free);
122 static inline int copy_inodes(struct btrfs_convert_context *cctx,
123 struct btrfs_root *root, int datacsum,
124 int packing, int noxattr, struct task_ctx *p)
126 return cctx->convert_ops->copy_inodes(cctx, root, datacsum, packing,
130 static inline void convert_close_fs(struct btrfs_convert_context *cctx)
132 cctx->convert_ops->close_fs(cctx);
135 static int intersect_with_sb(u64 bytenr, u64 num_bytes)
140 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
141 offset = btrfs_sb_offset(i);
142 offset &= ~((u64)BTRFS_STRIPE_LEN - 1);
144 if (bytenr < offset + BTRFS_STRIPE_LEN &&
145 bytenr + num_bytes > offset)
151 static int convert_insert_dirent(struct btrfs_trans_handle *trans,
152 struct btrfs_root *root,
153 const char *name, size_t name_len,
154 u64 dir, u64 objectid,
155 u8 file_type, u64 index_cnt,
156 struct btrfs_inode_item *inode)
160 struct btrfs_key location = {
161 .objectid = objectid,
163 .type = BTRFS_INODE_ITEM_KEY,
166 ret = btrfs_insert_dir_item(trans, root, name, name_len,
167 dir, &location, file_type, index_cnt);
170 ret = btrfs_insert_inode_ref(trans, root, name, name_len,
171 objectid, dir, index_cnt);
174 inode_size = btrfs_stack_inode_size(inode) + name_len * 2;
175 btrfs_set_stack_inode_size(inode, inode_size);
180 static int read_disk_extent(struct btrfs_root *root, u64 bytenr,
181 u32 num_bytes, char *buffer)
184 struct btrfs_fs_devices *fs_devs = root->fs_info->fs_devices;
186 ret = pread(fs_devs->latest_bdev, buffer, num_bytes, bytenr);
187 if (ret != num_bytes)
196 static int csum_disk_extent(struct btrfs_trans_handle *trans,
197 struct btrfs_root *root,
198 u64 disk_bytenr, u64 num_bytes)
200 u32 blocksize = root->sectorsize;
205 buffer = malloc(blocksize);
208 for (offset = 0; offset < num_bytes; offset += blocksize) {
209 ret = read_disk_extent(root, disk_bytenr + offset,
213 ret = btrfs_csum_file_block(trans,
214 root->fs_info->csum_root,
215 disk_bytenr + num_bytes,
216 disk_bytenr + offset,
225 struct blk_iterate_data {
226 struct btrfs_trans_handle *trans;
227 struct btrfs_root *root;
228 struct btrfs_root *convert_root;
229 struct btrfs_inode_item *inode;
240 static void init_blk_iterate_data(struct blk_iterate_data *data,
241 struct btrfs_trans_handle *trans,
242 struct btrfs_root *root,
243 struct btrfs_inode_item *inode,
244 u64 objectid, int checksum)
246 struct btrfs_key key;
251 data->objectid = objectid;
252 data->first_block = 0;
253 data->disk_block = 0;
254 data->num_blocks = 0;
255 data->boundary = (u64)-1;
256 data->checksum = checksum;
259 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
260 key.type = BTRFS_ROOT_ITEM_KEY;
261 key.offset = (u64)-1;
262 data->convert_root = btrfs_read_fs_root(root->fs_info, &key);
263 /* Impossible as we just opened it before */
264 BUG_ON(!data->convert_root || IS_ERR(data->convert_root));
265 data->convert_ino = BTRFS_FIRST_FREE_OBJECTID + 1;
269 * Record a file extent in original filesystem into btrfs one.
270 * The special point is, old disk_block can point to a reserved range.
271 * So here, we don't use disk_block directly but search convert_root
272 * to get the real disk_bytenr.
274 static int record_file_blocks(struct blk_iterate_data *data,
275 u64 file_block, u64 disk_block, u64 num_blocks)
278 struct btrfs_root *root = data->root;
279 struct btrfs_root *convert_root = data->convert_root;
280 struct btrfs_path *path;
281 u64 file_pos = file_block * root->sectorsize;
282 u64 old_disk_bytenr = disk_block * root->sectorsize;
283 u64 num_bytes = num_blocks * root->sectorsize;
284 u64 cur_off = old_disk_bytenr;
286 /* Hole, pass it to record_file_extent directly */
287 if (old_disk_bytenr == 0)
288 return btrfs_record_file_extent(data->trans, root,
289 data->objectid, data->inode, file_pos, 0,
292 path = btrfs_alloc_path();
297 * Search real disk bytenr from convert root
299 while (cur_off < old_disk_bytenr + num_bytes) {
300 struct btrfs_key key;
301 struct btrfs_file_extent_item *fi;
302 struct extent_buffer *node;
304 u64 extent_disk_bytenr;
305 u64 extent_num_bytes;
306 u64 real_disk_bytenr;
309 key.objectid = data->convert_ino;
310 key.type = BTRFS_EXTENT_DATA_KEY;
311 key.offset = cur_off;
313 ret = btrfs_search_slot(NULL, convert_root, &key, path, 0, 0);
317 ret = btrfs_previous_item(convert_root, path,
319 BTRFS_EXTENT_DATA_KEY);
327 node = path->nodes[0];
328 slot = path->slots[0];
329 btrfs_item_key_to_cpu(node, &key, slot);
330 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY ||
331 key.objectid != data->convert_ino ||
332 key.offset > cur_off);
333 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
334 extent_disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
335 extent_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
336 BUG_ON(cur_off - key.offset >= extent_num_bytes);
337 btrfs_release_path(path);
339 if (extent_disk_bytenr)
340 real_disk_bytenr = cur_off - key.offset +
343 real_disk_bytenr = 0;
344 cur_len = min(key.offset + extent_num_bytes,
345 old_disk_bytenr + num_bytes) - cur_off;
346 ret = btrfs_record_file_extent(data->trans, data->root,
347 data->objectid, data->inode, file_pos,
348 real_disk_bytenr, cur_len);
355 * No need to care about csum
356 * As every byte of old fs image is calculated for csum, no
357 * need to waste CPU cycles now.
360 btrfs_free_path(path);
364 static int block_iterate_proc(u64 disk_block, u64 file_block,
365 struct blk_iterate_data *idata)
370 struct btrfs_root *root = idata->root;
371 struct btrfs_block_group_cache *cache;
372 u64 bytenr = disk_block * root->sectorsize;
374 sb_region = intersect_with_sb(bytenr, root->sectorsize);
375 do_barrier = sb_region || disk_block >= idata->boundary;
376 if ((idata->num_blocks > 0 && do_barrier) ||
377 (file_block > idata->first_block + idata->num_blocks) ||
378 (disk_block != idata->disk_block + idata->num_blocks)) {
379 if (idata->num_blocks > 0) {
380 ret = record_file_blocks(idata, idata->first_block,
385 idata->first_block += idata->num_blocks;
386 idata->num_blocks = 0;
388 if (file_block > idata->first_block) {
389 ret = record_file_blocks(idata, idata->first_block,
390 0, file_block - idata->first_block);
396 bytenr += BTRFS_STRIPE_LEN - 1;
397 bytenr &= ~((u64)BTRFS_STRIPE_LEN - 1);
399 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
401 bytenr = cache->key.objectid + cache->key.offset;
404 idata->first_block = file_block;
405 idata->disk_block = disk_block;
406 idata->boundary = bytenr / root->sectorsize;
413 static int create_image_file_range(struct btrfs_trans_handle *trans,
414 struct btrfs_root *root,
415 struct cache_tree *used,
416 struct btrfs_inode_item *inode,
417 u64 ino, u64 bytenr, u64 *ret_len,
420 struct cache_extent *cache;
421 struct btrfs_block_group_cache *bg_cache;
427 if (bytenr != round_down(bytenr, root->sectorsize)) {
428 error("bytenr not sectorsize aligned: %llu",
429 (unsigned long long)bytenr);
432 if (len != round_down(len, root->sectorsize)) {
433 error("length not sectorsize aligned: %llu",
434 (unsigned long long)len);
437 len = min_t(u64, len, BTRFS_MAX_EXTENT_SIZE);
440 * Skip sb ranges first
441 * [0, 1M), [sb_offset(1), +64K), [sb_offset(2), +64K].
443 * Or we will insert a hole into current image file, and later
444 * migrate block will fail as there is already a file extent.
446 if (bytenr < 1024 * 1024) {
447 *ret_len = 1024 * 1024 - bytenr;
450 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
451 u64 cur = btrfs_sb_offset(i);
453 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
454 *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
458 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
459 u64 cur = btrfs_sb_offset(i);
464 * May still need to go through file extent inserts
466 if (bytenr < cur && bytenr + len >= cur) {
467 len = min_t(u64, len, cur - bytenr);
473 * Drop out, no need to insert anything
475 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
476 *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
481 cache = search_cache_extent(used, bytenr);
483 if (cache->start <= bytenr) {
485 * |///////Used///////|
489 len = min_t(u64, len, cache->start + cache->size -
491 disk_bytenr = bytenr;
498 len = min(len, cache->start - bytenr);
513 /* Check if the range is in a data block group */
514 bg_cache = btrfs_lookup_block_group(root->fs_info, bytenr);
517 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
520 /* The extent should never cross block group boundary */
521 len = min_t(u64, len, bg_cache->key.objectid +
522 bg_cache->key.offset - bytenr);
525 if (len != round_down(len, root->sectorsize)) {
526 error("remaining length not sectorsize aligned: %llu",
527 (unsigned long long)len);
530 ret = btrfs_record_file_extent(trans, root, ino, inode, bytenr,
536 ret = csum_disk_extent(trans, root, bytenr, len);
542 * Relocate old fs data in one reserved ranges
544 * Since all old fs data in reserved range is not covered by any chunk nor
545 * data extent, we don't need to handle any reference but add new
546 * extent/reference, which makes codes more clear
548 static int migrate_one_reserved_range(struct btrfs_trans_handle *trans,
549 struct btrfs_root *root,
550 struct cache_tree *used,
551 struct btrfs_inode_item *inode, int fd,
552 u64 ino, u64 start, u64 len, int datacsum)
556 u64 hole_start = start;
558 struct cache_extent *cache;
559 struct btrfs_key key;
560 struct extent_buffer *eb;
563 while (cur_off < start + len) {
564 cache = lookup_cache_extent(used, cur_off, cur_len);
567 cur_off = max(cache->start, cur_off);
568 cur_len = min(cache->start + cache->size, start + len) -
570 BUG_ON(cur_len < root->sectorsize);
572 /* reserve extent for the data */
573 ret = btrfs_reserve_extent(trans, root, cur_len, 0, 0, (u64)-1,
578 eb = malloc(sizeof(*eb) + cur_len);
584 ret = pread(fd, eb->data, cur_len, cur_off);
586 ret = (ret < 0 ? ret : -EIO);
590 eb->start = key.objectid;
591 eb->len = key.offset;
594 ret = write_and_map_eb(trans, root, eb);
599 /* Now handle extent item and file extent things */
600 ret = btrfs_record_file_extent(trans, root, ino, inode, cur_off,
601 key.objectid, key.offset);
604 /* Finally, insert csum items */
606 ret = csum_disk_extent(trans, root, key.objectid,
609 /* Don't forget to insert hole */
610 hole_len = cur_off - hole_start;
612 ret = btrfs_record_file_extent(trans, root, ino, inode,
613 hole_start, 0, hole_len);
618 cur_off += key.offset;
619 hole_start = cur_off;
620 cur_len = start + len - cur_off;
623 if (start + len - hole_start > 0)
624 ret = btrfs_record_file_extent(trans, root, ino, inode,
625 hole_start, 0, start + len - hole_start);
630 * Relocate the used ext2 data in reserved ranges
632 * [btrfs_sb_offset(1), +BTRFS_STRIPE_LEN)
633 * [btrfs_sb_offset(2), +BTRFS_STRIPE_LEN)
635 static int migrate_reserved_ranges(struct btrfs_trans_handle *trans,
636 struct btrfs_root *root,
637 struct cache_tree *used,
638 struct btrfs_inode_item *inode, int fd,
639 u64 ino, u64 total_bytes, int datacsum)
647 cur_len = 1024 * 1024;
648 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
649 cur_off, cur_len, datacsum);
653 /* second sb(fisrt sb is included in 0~1M) */
654 cur_off = btrfs_sb_offset(1);
655 cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
656 if (cur_off > total_bytes)
658 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
659 cur_off, cur_len, datacsum);
664 cur_off = btrfs_sb_offset(2);
665 cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
666 if (cur_off > total_bytes)
668 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
669 cur_off, cur_len, datacsum);
674 * Helper for expand and merge extent_cache for wipe_one_reserved_range() to
675 * handle wiping a range that exists in cache.
677 static int _expand_extent_cache(struct cache_tree *tree,
678 struct cache_extent *entry,
679 u64 min_stripe_size, int backward)
681 struct cache_extent *ce;
684 if (entry->size >= min_stripe_size)
686 diff = min_stripe_size - entry->size;
689 ce = prev_cache_extent(entry);
692 if (ce->start + ce->size >= entry->start - diff) {
693 /* Directly merge with previous extent */
694 ce->size = entry->start + entry->size - ce->start;
695 remove_cache_extent(tree, entry);
700 /* No overlap, normal extent */
701 if (entry->start < diff) {
702 error("cannot find space for data chunk layout");
705 entry->start -= diff;
709 ce = next_cache_extent(entry);
712 if (entry->start + entry->size + diff >= ce->start) {
713 /* Directly merge with next extent */
714 entry->size = ce->start + ce->size - entry->start;
715 remove_cache_extent(tree, ce);
725 * Remove one reserve range from given cache tree
726 * if min_stripe_size is non-zero, it will ensure for split case,
727 * all its split cache extent is no smaller than @min_strip_size / 2.
729 static int wipe_one_reserved_range(struct cache_tree *tree,
730 u64 start, u64 len, u64 min_stripe_size,
733 struct cache_extent *cache;
736 BUG_ON(ensure_size && min_stripe_size == 0);
738 * The logical here is simplified to handle special cases only
739 * So we don't need to consider merge case for ensure_size
741 BUG_ON(min_stripe_size && (min_stripe_size < len * 2 ||
742 min_stripe_size / 2 < BTRFS_STRIPE_LEN));
744 /* Also, wipe range should already be aligned */
745 BUG_ON(start != round_down(start, BTRFS_STRIPE_LEN) ||
746 start + len != round_up(start + len, BTRFS_STRIPE_LEN));
748 min_stripe_size /= 2;
750 cache = lookup_cache_extent(tree, start, len);
754 if (start <= cache->start) {
756 * |--------cache---------|
759 BUG_ON(start + len <= cache->start);
762 * The wipe size is smaller than min_stripe_size / 2,
763 * so the result length should still meet min_stripe_size
764 * And no need to do alignment
766 cache->size -= (start + len - cache->start);
767 if (cache->size == 0) {
768 remove_cache_extent(tree, cache);
773 BUG_ON(ensure_size && cache->size < min_stripe_size);
775 cache->start = start + len;
777 } else if (start > cache->start && start + len < cache->start +
780 * |-------cache-----|
783 u64 old_start = cache->start;
784 u64 old_len = cache->size;
785 u64 insert_start = start + len;
788 cache->size = start - cache->start;
789 /* Expand the leading half part if needed */
790 if (ensure_size && cache->size < min_stripe_size) {
791 ret = _expand_extent_cache(tree, cache,
797 /* And insert the new one */
798 insert_len = old_start + old_len - start - len;
799 ret = add_merge_cache_extent(tree, insert_start, insert_len);
803 /* Expand the last half part if needed */
804 if (ensure_size && insert_len < min_stripe_size) {
805 cache = lookup_cache_extent(tree, insert_start,
807 if (!cache || cache->start != insert_start ||
808 cache->size != insert_len)
810 ret = _expand_extent_cache(tree, cache,
819 * Wipe len should be small enough and no need to expand the
822 cache->size = start - cache->start;
823 BUG_ON(ensure_size && cache->size < min_stripe_size);
828 * Remove reserved ranges from given cache_tree
830 * It will remove the following ranges
832 * 2) 2nd superblock, +64K (make sure chunks are 64K aligned)
833 * 3) 3rd superblock, +64K
835 * @min_stripe must be given for safety check
836 * and if @ensure_size is given, it will ensure affected cache_extent will be
837 * larger than min_stripe_size
839 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
844 ret = wipe_one_reserved_range(tree, 0, 1024 * 1024, min_stripe_size,
848 ret = wipe_one_reserved_range(tree, btrfs_sb_offset(1),
849 BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
852 ret = wipe_one_reserved_range(tree, btrfs_sb_offset(2),
853 BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
857 static int calculate_available_space(struct btrfs_convert_context *cctx)
859 struct cache_tree *used = &cctx->used;
860 struct cache_tree *data_chunks = &cctx->data_chunks;
861 struct cache_tree *free = &cctx->free;
862 struct cache_extent *cache;
865 * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
866 * works without need to consider overlap
868 u64 min_stripe_size = 2 * 16 * 1024 * 1024;
871 /* Calculate data_chunks */
872 for (cache = first_cache_extent(used); cache;
873 cache = next_cache_extent(cache)) {
876 if (cache->start + cache->size < cur_off)
878 if (cache->start > cur_off + min_stripe_size)
879 cur_off = cache->start;
880 cur_len = max(cache->start + cache->size - cur_off,
882 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
888 * remove reserved ranges, so we won't ever bother relocating an old
889 * filesystem extent to other place.
891 ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
897 * Calculate free space
898 * Always round up the start bytenr, to avoid metadata extent corss
899 * stripe boundary, as later mkfs_convert() won't have all the extent
902 for (cache = first_cache_extent(data_chunks); cache;
903 cache = next_cache_extent(cache)) {
904 if (cache->start < cur_off)
906 if (cache->start > cur_off) {
910 len = cache->start - round_up(cur_off,
912 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
914 ret = add_merge_cache_extent(free, insert_start, len);
918 cur_off = cache->start + cache->size;
920 /* Don't forget the last range */
921 if (cctx->total_bytes > cur_off) {
922 u64 len = cctx->total_bytes - cur_off;
925 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
927 ret = add_merge_cache_extent(free, insert_start, len);
932 /* Remove reserved bytes */
933 ret = wipe_reserved_ranges(free, min_stripe_size, 0);
939 * Read used space, and since we have the used space,
940 * calcuate data_chunks and free for later mkfs
942 static int convert_read_used_space(struct btrfs_convert_context *cctx)
946 ret = cctx->convert_ops->read_used_space(cctx);
950 ret = calculate_available_space(cctx);
955 * Create the fs image file of old filesystem.
957 * This is completely fs independent as we have cctx->used, only
958 * need to create file extents pointing to all the positions.
960 static int create_image(struct btrfs_root *root,
961 struct btrfs_mkfs_config *cfg,
962 struct btrfs_convert_context *cctx, int fd,
963 u64 size, char *name, int datacsum)
965 struct btrfs_inode_item buf;
966 struct btrfs_trans_handle *trans;
967 struct btrfs_path *path = NULL;
968 struct btrfs_key key;
969 struct cache_extent *cache;
970 struct cache_tree used_tmp;
973 u64 flags = BTRFS_INODE_READONLY;
977 flags |= BTRFS_INODE_NODATASUM;
979 trans = btrfs_start_transaction(root, 1);
983 cache_tree_init(&used_tmp);
985 ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
989 ret = btrfs_new_inode(trans, root, ino, 0400 | S_IFREG);
992 ret = btrfs_change_inode_flags(trans, root, ino, flags);
995 ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
996 strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
1000 path = btrfs_alloc_path();
1006 key.type = BTRFS_INODE_ITEM_KEY;
1009 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1011 ret = (ret > 0 ? -ENOENT : ret);
1014 read_extent_buffer(path->nodes[0], &buf,
1015 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1017 btrfs_release_path(path);
1020 * Create a new used space cache, which doesn't contain the reserved
1023 for (cache = first_cache_extent(&cctx->used); cache;
1024 cache = next_cache_extent(cache)) {
1025 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1029 ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1034 * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1035 * can't handle bytenr 0(will consider it as a hole)
1038 while (cur < size) {
1039 u64 len = size - cur;
1041 ret = create_image_file_range(trans, root, &used_tmp,
1042 &buf, ino, cur, &len, datacsum);
1047 /* Handle the reserved ranges */
1048 ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1049 cfg->num_bytes, datacsum);
1053 key.type = BTRFS_INODE_ITEM_KEY;
1055 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1057 ret = (ret > 0 ? -ENOENT : ret);
1060 btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1061 write_extent_buffer(path->nodes[0], &buf,
1062 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1065 free_extent_cache_tree(&used_tmp);
1066 btrfs_free_path(path);
1067 btrfs_commit_transaction(trans, root);
1071 static struct btrfs_root* link_subvol(struct btrfs_root *root,
1072 const char *base, u64 root_objectid)
1074 struct btrfs_trans_handle *trans;
1075 struct btrfs_fs_info *fs_info = root->fs_info;
1076 struct btrfs_root *tree_root = fs_info->tree_root;
1077 struct btrfs_root *new_root = NULL;
1078 struct btrfs_path *path;
1079 struct btrfs_inode_item *inode_item;
1080 struct extent_buffer *leaf;
1081 struct btrfs_key key;
1082 u64 dirid = btrfs_root_dirid(&root->root_item);
1084 char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1090 if (len == 0 || len > BTRFS_NAME_LEN)
1093 path = btrfs_alloc_path();
1097 key.objectid = dirid;
1098 key.type = BTRFS_DIR_INDEX_KEY;
1099 key.offset = (u64)-1;
1101 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1103 error("search for DIR_INDEX dirid %llu failed: %d",
1104 (unsigned long long)dirid, ret);
1108 if (path->slots[0] > 0) {
1110 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1111 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1112 index = key.offset + 1;
1114 btrfs_release_path(path);
1116 trans = btrfs_start_transaction(root, 1);
1118 error("unable to start transaction");
1122 key.objectid = dirid;
1124 key.type = BTRFS_INODE_ITEM_KEY;
1126 ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1128 error("search for INODE_ITEM %llu failed: %d",
1129 (unsigned long long)dirid, ret);
1132 leaf = path->nodes[0];
1133 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1134 struct btrfs_inode_item);
1136 key.objectid = root_objectid;
1137 key.offset = (u64)-1;
1138 key.type = BTRFS_ROOT_ITEM_KEY;
1140 memcpy(buf, base, len);
1141 for (i = 0; i < 1024; i++) {
1142 ret = btrfs_insert_dir_item(trans, root, buf, len,
1143 dirid, &key, BTRFS_FT_DIR, index);
1146 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1147 if (len < 1 || len > BTRFS_NAME_LEN) {
1155 btrfs_set_inode_size(leaf, inode_item, len * 2 +
1156 btrfs_inode_size(leaf, inode_item));
1157 btrfs_mark_buffer_dirty(leaf);
1158 btrfs_release_path(path);
1160 /* add the backref first */
1161 ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1162 BTRFS_ROOT_BACKREF_KEY,
1163 root->root_key.objectid,
1164 dirid, index, buf, len);
1166 error("unable to add root backref for %llu: %d",
1167 root->root_key.objectid, ret);
1171 /* now add the forward ref */
1172 ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1173 BTRFS_ROOT_REF_KEY, root_objectid,
1174 dirid, index, buf, len);
1176 error("unable to add root ref for %llu: %d",
1177 root->root_key.objectid, ret);
1181 ret = btrfs_commit_transaction(trans, root);
1183 error("transaction commit failed: %d", ret);
1187 new_root = btrfs_read_fs_root(fs_info, &key);
1188 if (IS_ERR(new_root)) {
1189 error("unable to fs read root: %lu", PTR_ERR(new_root));
1193 btrfs_free_path(path);
1197 static int create_subvol(struct btrfs_trans_handle *trans,
1198 struct btrfs_root *root, u64 root_objectid)
1200 struct extent_buffer *tmp;
1201 struct btrfs_root *new_root;
1202 struct btrfs_key key;
1203 struct btrfs_root_item root_item;
1206 ret = btrfs_copy_root(trans, root, root->node, &tmp,
1211 memcpy(&root_item, &root->root_item, sizeof(root_item));
1212 btrfs_set_root_bytenr(&root_item, tmp->start);
1213 btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
1214 btrfs_set_root_generation(&root_item, trans->transid);
1215 free_extent_buffer(tmp);
1217 key.objectid = root_objectid;
1218 key.type = BTRFS_ROOT_ITEM_KEY;
1219 key.offset = trans->transid;
1220 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1223 key.offset = (u64)-1;
1224 new_root = btrfs_read_fs_root(root->fs_info, &key);
1225 if (!new_root || IS_ERR(new_root)) {
1226 error("unable to fs read root: %lu", PTR_ERR(new_root));
1227 return PTR_ERR(new_root);
1230 ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
1236 * New make_btrfs() has handle system and meta chunks quite well.
1237 * So only need to add remaining data chunks.
1239 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
1240 struct btrfs_fs_info *fs_info,
1241 struct btrfs_mkfs_config *cfg,
1242 struct btrfs_convert_context *cctx)
1244 struct btrfs_root *extent_root = fs_info->extent_root;
1245 struct cache_tree *data_chunks = &cctx->data_chunks;
1246 struct cache_extent *cache;
1251 * Don't create data chunk over 10% of the convert device
1252 * And for single chunk, don't create chunk larger than 1G.
1254 max_chunk_size = cfg->num_bytes / 10;
1255 max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
1256 max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
1258 for (cache = first_cache_extent(data_chunks); cache;
1259 cache = next_cache_extent(cache)) {
1260 u64 cur = cache->start;
1262 while (cur < cache->start + cache->size) {
1264 u64 cur_backup = cur;
1266 len = min(max_chunk_size,
1267 cache->start + cache->size - cur);
1268 ret = btrfs_alloc_data_chunk(trans, extent_root,
1270 BTRFS_BLOCK_GROUP_DATA, 1);
1273 ret = btrfs_make_block_group(trans, extent_root, 0,
1274 BTRFS_BLOCK_GROUP_DATA,
1275 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1286 * Init the temp btrfs to a operational status.
1288 * It will fix the extent usage accounting(XXX: Do we really need?) and
1289 * insert needed data chunks, to ensure all old fs data extents are covered
1290 * by DATA chunks, preventing wrong chunks are allocated.
1292 * And also create convert image subvolume and relocation tree.
1293 * (XXX: Not need again?)
1294 * But the convert image subvolume is *NOT* linked to fs tree yet.
1296 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
1297 struct btrfs_convert_context *cctx, int datacsum,
1298 int packing, int noxattr)
1300 struct btrfs_key location;
1301 struct btrfs_trans_handle *trans;
1302 struct btrfs_fs_info *fs_info = root->fs_info;
1306 * Don't alloc any metadata/system chunk, as we don't want
1307 * any meta/sys chunk allcated before all data chunks are inserted.
1308 * Or we screw up the chunk layout just like the old implement.
1310 fs_info->avoid_sys_chunk_alloc = 1;
1311 fs_info->avoid_meta_chunk_alloc = 1;
1312 trans = btrfs_start_transaction(root, 1);
1314 error("unable to start transaction");
1318 ret = btrfs_fix_block_accounting(trans, root);
1321 ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
1324 ret = btrfs_make_root_dir(trans, fs_info->tree_root,
1325 BTRFS_ROOT_TREE_DIR_OBJECTID);
1328 memcpy(&location, &root->root_key, sizeof(location));
1329 location.offset = (u64)-1;
1330 ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
1331 btrfs_super_root_dir(fs_info->super_copy),
1332 &location, BTRFS_FT_DIR, 0);
1335 ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
1337 btrfs_super_root_dir(fs_info->super_copy), 0);
1340 btrfs_set_root_dirid(&fs_info->fs_root->root_item,
1341 BTRFS_FIRST_FREE_OBJECTID);
1343 /* subvol for fs image file */
1344 ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
1346 error("failed to create subvolume image root: %d", ret);
1349 /* subvol for data relocation tree */
1350 ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1352 error("failed to create DATA_RELOC root: %d", ret);
1356 ret = btrfs_commit_transaction(trans, root);
1357 fs_info->avoid_sys_chunk_alloc = 0;
1358 fs_info->avoid_meta_chunk_alloc = 0;
1364 * Migrate super block to its default position and zero 0 ~ 16k
1366 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
1369 struct extent_buffer *buf;
1370 struct btrfs_super_block *super;
1374 buf = malloc(sizeof(*buf) + sectorsize);
1378 buf->len = sectorsize;
1379 ret = pread(fd, buf->data, sectorsize, old_bytenr);
1380 if (ret != sectorsize)
1383 super = (struct btrfs_super_block *)buf->data;
1384 BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
1385 btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
1387 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1388 ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
1389 if (ret != sectorsize)
1396 memset(buf->data, 0, sectorsize);
1397 for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
1398 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
1399 if (len > sectorsize)
1401 ret = pwrite(fd, buf->data, len, bytenr);
1403 fprintf(stderr, "unable to zero fill device\n");
1417 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
1419 struct btrfs_chunk *chunk;
1420 struct btrfs_disk_key *key;
1421 u32 sectorsize = btrfs_super_sectorsize(super);
1423 key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1424 chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1425 sizeof(struct btrfs_disk_key));
1427 btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1428 btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1429 btrfs_set_disk_key_offset(key, 0);
1431 btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
1432 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1433 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1434 btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1435 btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1436 btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1437 btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1438 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1439 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1440 chunk->stripe.devid = super->dev_item.devid;
1441 btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1442 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1443 btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1447 #if BTRFSCONVERT_EXT2
1450 * Open Ext2fs in readonly mode, read block allocation bitmap and
1451 * inode bitmap into memory.
1453 static int ext2_open_fs(struct btrfs_convert_context *cctx, const char *name)
1456 ext2_filsys ext2_fs;
1460 ret = ext2fs_open(name, 0, 0, 0, unix_io_manager, &ext2_fs);
1462 fprintf(stderr, "ext2fs_open: %s\n", error_message(ret));
1466 * We need to know exactly the used space, some RO compat flags like
1467 * BIGALLOC will affect how used space is present.
1468 * So we need manuall check any unsupported RO compat flags
1470 ro_feature = ext2_fs->super->s_feature_ro_compat;
1471 if (ro_feature & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
1473 "unsupported RO features detected: %x, abort convert to avoid possible corruption",
1474 ro_feature & ~EXT2_LIB_FEATURE_COMPAT_SUPP);
1477 ret = ext2fs_read_inode_bitmap(ext2_fs);
1479 fprintf(stderr, "ext2fs_read_inode_bitmap: %s\n",
1480 error_message(ret));
1483 ret = ext2fs_read_block_bitmap(ext2_fs);
1485 fprintf(stderr, "ext2fs_read_block_bitmap: %s\n",
1486 error_message(ret));
1490 * search each block group for a free inode. this set up
1491 * uninit block/inode bitmaps appropriately.
1494 while (ino <= ext2_fs->super->s_inodes_count) {
1496 ext2fs_new_inode(ext2_fs, ino, 0, NULL, &foo);
1497 ino += EXT2_INODES_PER_GROUP(ext2_fs->super);
1500 if (!(ext2_fs->super->s_feature_incompat &
1501 EXT2_FEATURE_INCOMPAT_FILETYPE)) {
1502 fprintf(stderr, "filetype feature is missing\n");
1506 cctx->fs_data = ext2_fs;
1507 cctx->blocksize = ext2_fs->blocksize;
1508 cctx->block_count = ext2_fs->super->s_blocks_count;
1509 cctx->total_bytes = ext2_fs->blocksize * ext2_fs->super->s_blocks_count;
1510 cctx->volume_name = strndup(ext2_fs->super->s_volume_name, 16);
1511 cctx->first_data_block = ext2_fs->super->s_first_data_block;
1512 cctx->inodes_count = ext2_fs->super->s_inodes_count;
1513 cctx->free_inodes_count = ext2_fs->super->s_free_inodes_count;
1516 ext2fs_close(ext2_fs);
1520 static int __ext2_add_one_block(ext2_filsys fs, char *bitmap,
1521 unsigned long group_nr, struct cache_tree *used)
1523 unsigned long offset;
1527 offset = fs->super->s_first_data_block;
1528 offset /= EXT2FS_CLUSTER_RATIO(fs);
1529 offset += group_nr * EXT2_CLUSTERS_PER_GROUP(fs->super);
1530 for (i = 0; i < EXT2_CLUSTERS_PER_GROUP(fs->super); i++) {
1531 if (ext2fs_test_bit(i, bitmap)) {
1534 start = (i + offset) * EXT2FS_CLUSTER_RATIO(fs);
1535 start *= fs->blocksize;
1536 ret = add_merge_cache_extent(used, start,
1546 * Read all used ext2 space into cctx->used cache tree
1548 static int ext2_read_used_space(struct btrfs_convert_context *cctx)
1550 ext2_filsys fs = (ext2_filsys)cctx->fs_data;
1551 blk64_t blk_itr = EXT2FS_B2C(fs, fs->super->s_first_data_block);
1552 struct cache_tree *used_tree = &cctx->used;
1553 char *block_bitmap = NULL;
1558 block_nbytes = EXT2_CLUSTERS_PER_GROUP(fs->super) / 8;
1559 /* Shouldn't happen */
1560 BUG_ON(!fs->block_map);
1562 block_bitmap = malloc(block_nbytes);
1566 for (i = 0; i < fs->group_desc_count; i++) {
1567 ret = ext2fs_get_block_bitmap_range(fs->block_map, blk_itr,
1568 block_nbytes * 8, block_bitmap);
1570 error("fail to get bitmap from ext2, %s",
1574 ret = __ext2_add_one_block(fs, block_bitmap, i, used_tree);
1576 error("fail to build used space tree, %s",
1580 blk_itr += EXT2_CLUSTERS_PER_GROUP(fs->super);
1587 static void ext2_close_fs(struct btrfs_convert_context *cctx)
1589 if (cctx->volume_name) {
1590 free(cctx->volume_name);
1591 cctx->volume_name = NULL;
1593 ext2fs_close(cctx->fs_data);
1596 struct dir_iterate_data {
1597 struct btrfs_trans_handle *trans;
1598 struct btrfs_root *root;
1599 struct btrfs_inode_item *inode;
1606 static u8 ext2_filetype_conversion_table[EXT2_FT_MAX] = {
1607 [EXT2_FT_UNKNOWN] = BTRFS_FT_UNKNOWN,
1608 [EXT2_FT_REG_FILE] = BTRFS_FT_REG_FILE,
1609 [EXT2_FT_DIR] = BTRFS_FT_DIR,
1610 [EXT2_FT_CHRDEV] = BTRFS_FT_CHRDEV,
1611 [EXT2_FT_BLKDEV] = BTRFS_FT_BLKDEV,
1612 [EXT2_FT_FIFO] = BTRFS_FT_FIFO,
1613 [EXT2_FT_SOCK] = BTRFS_FT_SOCK,
1614 [EXT2_FT_SYMLINK] = BTRFS_FT_SYMLINK,
1617 static int ext2_dir_iterate_proc(ext2_ino_t dir, int entry,
1618 struct ext2_dir_entry *dirent,
1619 int offset, int blocksize,
1620 char *buf,void *priv_data)
1625 char dotdot[] = "..";
1626 struct dir_iterate_data *idata = (struct dir_iterate_data *)priv_data;
1629 name_len = dirent->name_len & 0xFF;
1631 objectid = dirent->inode + INO_OFFSET;
1632 if (!strncmp(dirent->name, dotdot, name_len)) {
1633 if (name_len == 2) {
1634 BUG_ON(idata->parent != 0);
1635 idata->parent = objectid;
1639 if (dirent->inode < EXT2_GOOD_OLD_FIRST_INO)
1642 file_type = dirent->name_len >> 8;
1643 BUG_ON(file_type > EXT2_FT_SYMLINK);
1645 ret = convert_insert_dirent(idata->trans, idata->root, dirent->name,
1646 name_len, idata->objectid, objectid,
1647 ext2_filetype_conversion_table[file_type],
1648 idata->index_cnt, idata->inode);
1650 idata->errcode = ret;
1658 static int ext2_create_dir_entries(struct btrfs_trans_handle *trans,
1659 struct btrfs_root *root, u64 objectid,
1660 struct btrfs_inode_item *btrfs_inode,
1661 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
1665 struct dir_iterate_data data = {
1668 .inode = btrfs_inode,
1669 .objectid = objectid,
1675 err = ext2fs_dir_iterate2(ext2_fs, ext2_ino, 0, NULL,
1676 ext2_dir_iterate_proc, &data);
1680 if (ret == 0 && data.parent == objectid) {
1681 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
1682 objectid, objectid, 0);
1686 fprintf(stderr, "ext2fs_dir_iterate2: %s\n", error_message(err));
1690 static int ext2_block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
1691 e2_blkcnt_t blockcnt, blk_t ref_block,
1692 int ref_offset, void *priv_data)
1695 struct blk_iterate_data *idata;
1696 idata = (struct blk_iterate_data *)priv_data;
1697 ret = block_iterate_proc(*blocknr, blockcnt, idata);
1699 idata->errcode = ret;
1706 * traverse file's data blocks, record these data blocks as file extents.
1708 static int ext2_create_file_extents(struct btrfs_trans_handle *trans,
1709 struct btrfs_root *root, u64 objectid,
1710 struct btrfs_inode_item *btrfs_inode,
1711 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1712 int datacsum, int packing)
1715 char *buffer = NULL;
1718 u32 sectorsize = root->sectorsize;
1719 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1720 struct blk_iterate_data data;
1722 init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
1725 err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
1726 NULL, ext2_block_iterate_proc, &data);
1732 if (packing && data.first_block == 0 && data.num_blocks > 0 &&
1733 inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
1734 u64 num_bytes = data.num_blocks * sectorsize;
1735 u64 disk_bytenr = data.disk_block * sectorsize;
1738 buffer = malloc(num_bytes);
1741 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
1744 if (num_bytes > inode_size)
1745 num_bytes = inode_size;
1746 ret = btrfs_insert_inline_extent(trans, root, objectid,
1747 0, buffer, num_bytes);
1750 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
1751 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
1752 } else if (data.num_blocks > 0) {
1753 ret = record_file_blocks(&data, data.first_block,
1754 data.disk_block, data.num_blocks);
1758 data.first_block += data.num_blocks;
1759 last_block = (inode_size + sectorsize - 1) / sectorsize;
1760 if (last_block > data.first_block) {
1761 ret = record_file_blocks(&data, data.first_block, 0,
1762 last_block - data.first_block);
1768 fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
1772 static int ext2_create_symbol_link(struct btrfs_trans_handle *trans,
1773 struct btrfs_root *root, u64 objectid,
1774 struct btrfs_inode_item *btrfs_inode,
1775 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1776 struct ext2_inode *ext2_inode)
1780 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1781 if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
1782 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
1783 ret = ext2_create_file_extents(trans, root, objectid,
1784 btrfs_inode, ext2_fs, ext2_ino, 1, 1);
1785 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
1789 pathname = (char *)&(ext2_inode->i_block[0]);
1790 BUG_ON(pathname[inode_size] != 0);
1791 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
1792 pathname, inode_size + 1);
1793 btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
1798 * Following xattr/acl related codes are based on codes in
1799 * fs/ext3/xattr.c and fs/ext3/acl.c
1801 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
1802 #define EXT2_XATTR_BFIRST(ptr) \
1803 ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
1804 #define EXT2_XATTR_IHDR(inode) \
1805 ((struct ext2_ext_attr_header *) ((void *)(inode) + \
1806 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
1807 #define EXT2_XATTR_IFIRST(inode) \
1808 ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
1809 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
1811 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
1814 struct ext2_ext_attr_entry *next;
1816 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1817 next = EXT2_EXT_ATTR_NEXT(entry);
1818 if ((void *)next >= end)
1825 static int ext2_xattr_check_block(const char *buf, size_t size)
1828 struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
1830 if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
1831 header->h_blocks != 1)
1833 error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
1837 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
1840 size_t value_size = entry->e_value_size;
1842 if (entry->e_value_block != 0 || value_size > size ||
1843 entry->e_value_offs + value_size > size)
1848 #define EXT2_ACL_VERSION 0x0001
1850 /* 23.2.5 acl_tag_t values */
1852 #define ACL_UNDEFINED_TAG (0x00)
1853 #define ACL_USER_OBJ (0x01)
1854 #define ACL_USER (0x02)
1855 #define ACL_GROUP_OBJ (0x04)
1856 #define ACL_GROUP (0x08)
1857 #define ACL_MASK (0x10)
1858 #define ACL_OTHER (0x20)
1860 /* 23.2.7 ACL qualifier constants */
1862 #define ACL_UNDEFINED_ID ((id_t)-1)
1873 } ext2_acl_entry_short;
1879 static inline int ext2_acl_count(size_t size)
1882 size -= sizeof(ext2_acl_header);
1883 s = size - 4 * sizeof(ext2_acl_entry_short);
1885 if (size % sizeof(ext2_acl_entry_short))
1887 return size / sizeof(ext2_acl_entry_short);
1889 if (s % sizeof(ext2_acl_entry))
1891 return s / sizeof(ext2_acl_entry) + 4;
1895 #define ACL_EA_VERSION 0x0002
1905 acl_ea_entry a_entries[0];
1908 static inline size_t acl_ea_size(int count)
1910 return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
1913 static int ext2_acl_to_xattr(void *dst, const void *src,
1914 size_t dst_size, size_t src_size)
1917 const void *end = src + src_size;
1918 acl_ea_header *ext_acl = (acl_ea_header *)dst;
1919 acl_ea_entry *dst_entry = ext_acl->a_entries;
1920 ext2_acl_entry *src_entry;
1922 if (src_size < sizeof(ext2_acl_header))
1924 if (((ext2_acl_header *)src)->a_version !=
1925 cpu_to_le32(EXT2_ACL_VERSION))
1927 src += sizeof(ext2_acl_header);
1928 count = ext2_acl_count(src_size);
1932 BUG_ON(dst_size < acl_ea_size(count));
1933 ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
1934 for (i = 0; i < count; i++, dst_entry++) {
1935 src_entry = (ext2_acl_entry *)src;
1936 if (src + sizeof(ext2_acl_entry_short) > end)
1938 dst_entry->e_tag = src_entry->e_tag;
1939 dst_entry->e_perm = src_entry->e_perm;
1940 switch (le16_to_cpu(src_entry->e_tag)) {
1945 src += sizeof(ext2_acl_entry_short);
1946 dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
1950 src += sizeof(ext2_acl_entry);
1953 dst_entry->e_id = src_entry->e_id;
1966 static char *xattr_prefix_table[] = {
1968 [2] = "system.posix_acl_access",
1969 [3] = "system.posix_acl_default",
1974 static int ext2_copy_single_xattr(struct btrfs_trans_handle *trans,
1975 struct btrfs_root *root, u64 objectid,
1976 struct ext2_ext_attr_entry *entry,
1977 const void *data, u32 datalen)
1982 void *databuf = NULL;
1983 char namebuf[XATTR_NAME_MAX + 1];
1985 name_index = entry->e_name_index;
1986 if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
1987 xattr_prefix_table[name_index] == NULL)
1989 name_len = strlen(xattr_prefix_table[name_index]) +
1991 if (name_len >= sizeof(namebuf))
1994 if (name_index == 2 || name_index == 3) {
1995 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
1996 databuf = malloc(bufsize);
1999 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
2005 strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
2006 strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
2007 if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
2008 sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
2009 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
2010 objectid - INO_OFFSET, name_len, namebuf);
2013 ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
2014 data, datalen, objectid);
2020 static int ext2_copy_extended_attrs(struct btrfs_trans_handle *trans,
2021 struct btrfs_root *root, u64 objectid,
2022 struct btrfs_inode_item *btrfs_inode,
2023 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
2029 u32 block_size = ext2_fs->blocksize;
2030 u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
2031 struct ext2_inode_large *ext2_inode;
2032 struct ext2_ext_attr_entry *entry;
2034 char *buffer = NULL;
2035 char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
2037 if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
2038 ext2_inode = (struct ext2_inode_large *)inode_buf;
2040 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
2044 err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
2047 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
2048 error_message(err));
2053 if (ext2_ino > ext2_fs->super->s_first_ino &&
2054 inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
2055 if (EXT2_GOOD_OLD_INODE_SIZE +
2056 ext2_inode->i_extra_isize > inode_size) {
2060 if (ext2_inode->i_extra_isize != 0 &&
2061 EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
2062 EXT2_EXT_ATTR_MAGIC) {
2068 void *end = (void *)ext2_inode + inode_size;
2069 entry = EXT2_XATTR_IFIRST(ext2_inode);
2070 total = end - (void *)entry;
2071 ret = ext2_xattr_check_names(entry, end);
2074 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
2075 ret = ext2_xattr_check_entry(entry, total);
2078 data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
2079 entry->e_value_offs;
2080 datalen = entry->e_value_size;
2081 ret = ext2_copy_single_xattr(trans, root, objectid,
2082 entry, data, datalen);
2085 entry = EXT2_EXT_ATTR_NEXT(entry);
2089 if (ext2_inode->i_file_acl == 0)
2092 buffer = malloc(block_size);
2097 err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
2099 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
2100 error_message(err));
2104 ret = ext2_xattr_check_block(buffer, block_size);
2108 entry = EXT2_XATTR_BFIRST(buffer);
2109 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
2110 ret = ext2_xattr_check_entry(entry, block_size);
2113 data = buffer + entry->e_value_offs;
2114 datalen = entry->e_value_size;
2115 ret = ext2_copy_single_xattr(trans, root, objectid,
2116 entry, data, datalen);
2119 entry = EXT2_EXT_ATTR_NEXT(entry);
2123 if ((void *)ext2_inode != inode_buf)
2127 #define MINORBITS 20
2128 #define MKDEV(ma, mi) (((ma) << MINORBITS) | (mi))
2130 static inline dev_t old_decode_dev(u16 val)
2132 return MKDEV((val >> 8) & 255, val & 255);
2135 static inline dev_t new_decode_dev(u32 dev)
2137 unsigned major = (dev & 0xfff00) >> 8;
2138 unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
2139 return MKDEV(major, minor);
2142 static void ext2_copy_inode_item(struct btrfs_inode_item *dst,
2143 struct ext2_inode *src, u32 blocksize)
2145 btrfs_set_stack_inode_generation(dst, 1);
2146 btrfs_set_stack_inode_sequence(dst, 0);
2147 btrfs_set_stack_inode_transid(dst, 1);
2148 btrfs_set_stack_inode_size(dst, src->i_size);
2149 btrfs_set_stack_inode_nbytes(dst, 0);
2150 btrfs_set_stack_inode_block_group(dst, 0);
2151 btrfs_set_stack_inode_nlink(dst, src->i_links_count);
2152 btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
2153 btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
2154 btrfs_set_stack_inode_mode(dst, src->i_mode);
2155 btrfs_set_stack_inode_rdev(dst, 0);
2156 btrfs_set_stack_inode_flags(dst, 0);
2157 btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
2158 btrfs_set_stack_timespec_nsec(&dst->atime, 0);
2159 btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
2160 btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
2161 btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
2162 btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
2163 btrfs_set_stack_timespec_sec(&dst->otime, 0);
2164 btrfs_set_stack_timespec_nsec(&dst->otime, 0);
2166 if (S_ISDIR(src->i_mode)) {
2167 btrfs_set_stack_inode_size(dst, 0);
2168 btrfs_set_stack_inode_nlink(dst, 1);
2170 if (S_ISREG(src->i_mode)) {
2171 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
2174 if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
2175 !S_ISLNK(src->i_mode)) {
2176 if (src->i_block[0]) {
2177 btrfs_set_stack_inode_rdev(dst,
2178 old_decode_dev(src->i_block[0]));
2180 btrfs_set_stack_inode_rdev(dst,
2181 new_decode_dev(src->i_block[1]));
2184 memset(&dst->reserved, 0, sizeof(dst->reserved));
2188 * copy a single inode. do all the required works, such as cloning
2189 * inode item, creating file extents and creating directory entries.
2191 static int ext2_copy_single_inode(struct btrfs_trans_handle *trans,
2192 struct btrfs_root *root, u64 objectid,
2193 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
2194 struct ext2_inode *ext2_inode,
2195 int datacsum, int packing, int noxattr)
2198 struct btrfs_inode_item btrfs_inode;
2200 if (ext2_inode->i_links_count == 0)
2203 ext2_copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
2204 if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
2205 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
2206 BTRFS_INODE_NODATASUM;
2207 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
2210 switch (ext2_inode->i_mode & S_IFMT) {
2212 ret = ext2_create_file_extents(trans, root, objectid,
2213 &btrfs_inode, ext2_fs, ext2_ino, datacsum, packing);
2216 ret = ext2_create_dir_entries(trans, root, objectid,
2217 &btrfs_inode, ext2_fs, ext2_ino);
2220 ret = ext2_create_symbol_link(trans, root, objectid,
2221 &btrfs_inode, ext2_fs, ext2_ino, ext2_inode);
2231 ret = ext2_copy_extended_attrs(trans, root, objectid,
2232 &btrfs_inode, ext2_fs, ext2_ino);
2236 return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
2240 * scan ext2's inode bitmap and copy all used inodes.
2242 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
2243 struct btrfs_root *root,
2244 int datacsum, int packing, int noxattr, struct task_ctx *p)
2246 ext2_filsys ext2_fs = cctx->fs_data;
2249 ext2_inode_scan ext2_scan;
2250 struct ext2_inode ext2_inode;
2251 ext2_ino_t ext2_ino;
2253 struct btrfs_trans_handle *trans;
2255 trans = btrfs_start_transaction(root, 1);
2258 err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
2260 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
2263 while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
2265 /* no more inodes */
2268 /* skip special inode in ext2fs */
2269 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
2270 ext2_ino != EXT2_ROOT_INO)
2272 objectid = ext2_ino + INO_OFFSET;
2273 ret = ext2_copy_single_inode(trans, root,
2274 objectid, ext2_fs, ext2_ino,
2275 &ext2_inode, datacsum, packing,
2277 p->cur_copy_inodes++;
2280 if (trans->blocks_used >= 4096) {
2281 ret = btrfs_commit_transaction(trans, root);
2283 trans = btrfs_start_transaction(root, 1);
2288 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
2291 ret = btrfs_commit_transaction(trans, root);
2293 ext2fs_close_inode_scan(ext2_scan);
2298 static const struct btrfs_convert_operations ext2_convert_ops = {
2300 .open_fs = ext2_open_fs,
2301 .read_used_space = ext2_read_used_space,
2302 .copy_inodes = ext2_copy_inodes,
2303 .close_fs = ext2_close_fs,
2308 static const struct btrfs_convert_operations *convert_operations[] = {
2309 #if BTRFSCONVERT_EXT2
2314 static int convert_open_fs(const char *devname,
2315 struct btrfs_convert_context *cctx)
2319 memset(cctx, 0, sizeof(*cctx));
2321 for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
2322 int ret = convert_operations[i]->open_fs(cctx, devname);
2325 cctx->convert_ops = convert_operations[i];
2330 fprintf(stderr, "No file system found to convert.\n");
2334 static int do_convert(const char *devname, int datacsum, int packing,
2335 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2336 int progress, u64 features)
2342 struct btrfs_root *root;
2343 struct btrfs_root *image_root;
2344 struct btrfs_convert_context cctx;
2345 struct btrfs_key key;
2346 char *subvol_name = NULL;
2347 struct task_ctx ctx;
2348 char features_buf[64];
2349 struct btrfs_mkfs_config mkfs_cfg;
2351 init_convert_context(&cctx);
2352 ret = convert_open_fs(devname, &cctx);
2355 ret = convert_read_used_space(&cctx);
2359 blocksize = cctx.blocksize;
2360 total_bytes = (u64)blocksize * (u64)cctx.block_count;
2361 if (blocksize < 4096) {
2362 error("block size is too small: %u < 4096", blocksize);
2365 if (btrfs_check_nodesize(nodesize, blocksize, features))
2367 fd = open(devname, O_RDWR);
2369 error("unable to open %s: %s", devname, strerror(errno));
2372 btrfs_parse_features_to_string(features_buf, features);
2373 if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2374 strcat(features_buf, " (default)");
2376 printf("create btrfs filesystem:\n");
2377 printf("\tblocksize: %u\n", blocksize);
2378 printf("\tnodesize: %u\n", nodesize);
2379 printf("\tfeatures: %s\n", features_buf);
2381 mkfs_cfg.label = cctx.volume_name;
2382 mkfs_cfg.num_bytes = total_bytes;
2383 mkfs_cfg.nodesize = nodesize;
2384 mkfs_cfg.sectorsize = blocksize;
2385 mkfs_cfg.stripesize = blocksize;
2386 mkfs_cfg.features = features;
2387 /* New convert need these space */
2388 memset(mkfs_cfg.chunk_uuid, 0, BTRFS_UUID_UNPARSED_SIZE);
2389 memset(mkfs_cfg.fs_uuid, 0, BTRFS_UUID_UNPARSED_SIZE);
2391 ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2393 error("unable to create initial ctree: %s", strerror(-ret));
2397 root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2398 OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2400 error("unable to open ctree");
2403 ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2405 error("unable to setup the root tree: %d", ret);
2409 printf("creating %s image file\n", cctx.convert_ops->name);
2410 ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2412 error("memory allocation failure for subvolume name: %s_saved",
2413 cctx.convert_ops->name);
2416 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2417 key.offset = (u64)-1;
2418 key.type = BTRFS_ROOT_ITEM_KEY;
2419 image_root = btrfs_read_fs_root(root->fs_info, &key);
2421 error("unable to create image subvolume");
2424 ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2425 mkfs_cfg.num_bytes, "image", datacsum);
2427 error("failed to create %s/image: %d", subvol_name, ret);
2431 printf("creating btrfs metadata");
2432 ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2433 ctx.cur_copy_inodes = 0;
2436 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2438 task_start(ctx.info);
2440 ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2442 error("error during copy_inodes %d", ret);
2446 task_stop(ctx.info);
2447 task_deinit(ctx.info);
2450 image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2452 error("unable to link subvolume %s", subvol_name);
2458 memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2459 if (copylabel == 1) {
2460 __strncpy_null(root->fs_info->super_copy->label,
2461 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2462 printf("copy label '%s'\n", root->fs_info->super_copy->label);
2463 } else if (copylabel == -1) {
2464 strcpy(root->fs_info->super_copy->label, fslabel);
2465 printf("set label to '%s'\n", fslabel);
2468 ret = close_ctree(root);
2470 error("close_ctree failed: %d", ret);
2473 convert_close_fs(&cctx);
2474 clean_convert_context(&cctx);
2477 * If this step succeed, we get a mountable btrfs. Otherwise
2478 * the source fs is left unchanged.
2480 ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2482 error("unable to migrate super block: %d", ret);
2486 root = open_ctree_fd(fd, devname, 0,
2487 OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2489 error("unable to open ctree for finalization");
2492 root->fs_info->finalize_on_close = 1;
2496 printf("conversion complete");
2499 clean_convert_context(&cctx);
2503 "an error occurred during conversion, filesystem is partially created but not finalized and not mountable");
2508 * Check if a non 1:1 mapped chunk can be rolled back.
2509 * For new convert, it's OK while for old convert it's not.
2511 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2513 struct btrfs_block_group_cache *bg;
2514 struct btrfs_key key;
2515 struct btrfs_path path;
2516 struct btrfs_root *extent_root = fs_info->extent_root;
2521 bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2524 bg_start = bg->key.objectid;
2525 bg_end = bg->key.objectid + bg->key.offset;
2527 key.objectid = bg_end;
2528 key.type = BTRFS_METADATA_ITEM_KEY;
2530 btrfs_init_path(&path);
2532 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2537 struct btrfs_extent_item *ei;
2539 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2547 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2548 if (key.type == BTRFS_METADATA_ITEM_KEY)
2550 /* Now it's EXTENT_ITEM_KEY only */
2551 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2552 struct btrfs_extent_item);
2554 * Found data extent, means this is old convert must follow 1:1
2557 if (btrfs_extent_flags(path.nodes[0], ei)
2558 & BTRFS_EXTENT_FLAG_DATA) {
2563 btrfs_release_path(&path);
2567 static int may_rollback(struct btrfs_root *root)
2569 struct btrfs_fs_info *info = root->fs_info;
2570 struct btrfs_multi_bio *multi = NULL;
2578 if (btrfs_super_num_devices(info->super_copy) != 1)
2581 bytenr = BTRFS_SUPER_INFO_OFFSET;
2582 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2585 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2586 &length, &multi, 0, NULL);
2588 if (ret == -ENOENT) {
2589 /* removed block group at the tail */
2590 if (length == (u64)-1)
2593 /* removed block group in the middle */
2599 num_stripes = multi->num_stripes;
2600 physical = multi->stripes[0].physical;
2603 if (num_stripes != 1) {
2604 error("num stripes for bytenr %llu is not 1", bytenr);
2609 * Extra check for new convert, as metadata chunk from new
2610 * convert is much more free than old convert, it doesn't need
2611 * to do 1:1 mapping.
2613 if (physical != bytenr) {
2615 * Check if it's a metadata chunk and has only metadata
2618 ret = may_rollback_chunk(info, bytenr);
2624 if (bytenr >= total_bytes)
2632 static int do_rollback(const char *devname)
2637 struct btrfs_root *root;
2638 struct btrfs_root *image_root;
2639 struct btrfs_root *chunk_root;
2640 struct btrfs_dir_item *dir;
2641 struct btrfs_inode_item *inode;
2642 struct btrfs_file_extent_item *fi;
2643 struct btrfs_trans_handle *trans;
2644 struct extent_buffer *leaf;
2645 struct btrfs_block_group_cache *cache1;
2646 struct btrfs_block_group_cache *cache2;
2647 struct btrfs_key key;
2648 struct btrfs_path path;
2649 struct extent_io_tree io_tree;
2664 extent_io_tree_init(&io_tree);
2666 fd = open(devname, O_RDWR);
2668 error("unable to open %s: %s", devname, strerror(errno));
2671 root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2673 error("unable to open ctree");
2676 ret = may_rollback(root);
2678 error("unable to do rollback: %d", ret);
2682 sectorsize = root->sectorsize;
2683 buf = malloc(sectorsize);
2685 error("unable to allocate memory");
2689 btrfs_init_path(&path);
2691 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2692 key.type = BTRFS_ROOT_BACKREF_KEY;
2693 key.offset = BTRFS_FS_TREE_OBJECTID;
2694 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2696 btrfs_release_path(&path);
2698 error("unable to convert ext2 image subvolume, is it deleted?");
2700 } else if (ret < 0) {
2701 error("unable to open ext2_saved, id %llu: %s",
2702 (unsigned long long)key.objectid, strerror(-ret));
2706 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2707 key.type = BTRFS_ROOT_ITEM_KEY;
2708 key.offset = (u64)-1;
2709 image_root = btrfs_read_fs_root(root->fs_info, &key);
2710 if (!image_root || IS_ERR(image_root)) {
2711 error("unable to open subvolume %llu: %ld",
2712 (unsigned long long)key.objectid, PTR_ERR(image_root));
2717 root_dir = btrfs_root_dirid(&root->root_item);
2718 dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2719 root_dir, name, strlen(name), 0);
2720 if (!dir || IS_ERR(dir)) {
2721 error("unable to find file %s: %ld", name, PTR_ERR(dir));
2724 leaf = path.nodes[0];
2725 btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2726 btrfs_release_path(&path);
2728 objectid = key.objectid;
2730 ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2732 error("unable to find inode item: %d", ret);
2735 leaf = path.nodes[0];
2736 inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2737 total_bytes = btrfs_inode_size(leaf, inode);
2738 btrfs_release_path(&path);
2740 key.objectid = objectid;
2742 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2743 ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2745 error("unable to find first file extent: %d", ret);
2746 btrfs_release_path(&path);
2750 /* build mapping tree for the relocated blocks */
2751 for (offset = 0; offset < total_bytes; ) {
2752 leaf = path.nodes[0];
2753 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2754 ret = btrfs_next_leaf(root, &path);
2760 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2761 if (key.objectid != objectid || key.offset != offset ||
2762 btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2765 fi = btrfs_item_ptr(leaf, path.slots[0],
2766 struct btrfs_file_extent_item);
2767 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2769 if (btrfs_file_extent_compression(leaf, fi) ||
2770 btrfs_file_extent_encryption(leaf, fi) ||
2771 btrfs_file_extent_other_encoding(leaf, fi))
2774 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2775 /* skip holes and direct mapped extents */
2776 if (bytenr == 0 || bytenr == offset)
2779 bytenr += btrfs_file_extent_offset(leaf, fi);
2780 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2782 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2783 cache2 = btrfs_lookup_block_group(root->fs_info,
2784 offset + num_bytes - 1);
2786 * Here we must take consideration of old and new convert
2788 * For old convert case, sign, there is no consist chunk type
2789 * that will cover the extent. META/DATA/SYS are all possible.
2790 * Just ensure relocate one is in SYS chunk.
2791 * For new convert case, they are all covered by DATA chunk.
2793 * So, there is not valid chunk type check for it now.
2795 if (cache1 != cache2)
2798 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2799 EXTENT_LOCKED, GFP_NOFS);
2800 set_state_private(&io_tree, offset, bytenr);
2802 offset += btrfs_file_extent_num_bytes(leaf, fi);
2805 btrfs_release_path(&path);
2807 if (offset < total_bytes) {
2808 error("unable to build extent mapping (offset %llu, total_bytes %llu)",
2809 (unsigned long long)offset,
2810 (unsigned long long)total_bytes);
2811 error("converted filesystem after balance is unable to rollback");
2815 first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2816 first_free &= ~((u64)sectorsize - 1);
2817 /* backup for extent #0 should exist */
2818 if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2819 error("no backup for the first extent");
2822 /* force no allocation from system block group */
2823 root->fs_info->system_allocs = -1;
2824 trans = btrfs_start_transaction(root, 1);
2827 * recow the whole chunk tree, this will remove all chunk tree blocks
2828 * from system block group
2830 chunk_root = root->fs_info->chunk_root;
2831 memset(&key, 0, sizeof(key));
2833 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2837 ret = btrfs_next_leaf(chunk_root, &path);
2841 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2842 btrfs_release_path(&path);
2844 btrfs_release_path(&path);
2849 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2853 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2854 num_bytes += btrfs_block_group_used(&cache1->item);
2856 offset = cache1->key.objectid + cache1->key.offset;
2858 /* only extent #0 left in system block group? */
2859 if (num_bytes > first_free) {
2861 "unable to empty system block group (num_bytes %llu, first_free %llu",
2862 (unsigned long long)num_bytes,
2863 (unsigned long long)first_free);
2866 /* create a system chunk that maps the whole device */
2867 ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2869 error("unable to update system chunk: %d", ret);
2873 ret = btrfs_commit_transaction(trans, root);
2876 ret = close_ctree(root);
2878 error("close_ctree failed: %d", ret);
2882 /* zero btrfs super block mirrors */
2883 memset(buf, 0, sectorsize);
2884 for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2885 bytenr = btrfs_sb_offset(i);
2886 if (bytenr >= total_bytes)
2888 ret = pwrite(fd, buf, sectorsize, bytenr);
2889 if (ret != sectorsize) {
2890 error("zeroing superblock mirror %d failed: %d",
2896 sb_bytenr = (u64)-1;
2897 /* copy all relocated blocks back */
2899 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2904 ret = get_state_private(&io_tree, start, &bytenr);
2907 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2910 while (start <= end) {
2911 if (start == BTRFS_SUPER_INFO_OFFSET) {
2915 ret = pread(fd, buf, sectorsize, bytenr);
2917 error("reading superblock at %llu failed: %d",
2918 (unsigned long long)bytenr, ret);
2921 BUG_ON(ret != sectorsize);
2922 ret = pwrite(fd, buf, sectorsize, start);
2924 error("writing superblock at %llu failed: %d",
2925 (unsigned long long)start, ret);
2928 BUG_ON(ret != sectorsize);
2930 start += sectorsize;
2931 bytenr += sectorsize;
2937 error("fsync failed: %s", strerror(errno));
2941 * finally, overwrite btrfs super block.
2943 ret = pread(fd, buf, sectorsize, sb_bytenr);
2945 error("reading primary superblock failed: %s",
2949 BUG_ON(ret != sectorsize);
2950 ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2952 error("writing primary superblock failed: %s",
2956 BUG_ON(ret != sectorsize);
2959 error("fsync failed: %s", strerror(errno));
2965 extent_io_tree_cleanup(&io_tree);
2966 printf("rollback complete\n");
2973 error("rollback aborted");
2977 static void print_usage(void)
2979 printf("usage: btrfs-convert [options] device\n");
2980 printf("options:\n");
2981 printf("\t-d|--no-datasum disable data checksum, sets NODATASUM\n");
2982 printf("\t-i|--no-xattr ignore xattrs and ACLs\n");
2983 printf("\t-n|--no-inline disable inlining of small files to metadata\n");
2984 printf("\t-N|--nodesize SIZE set filesystem metadata nodesize\n");
2985 printf("\t-r|--rollback roll back to the original filesystem\n");
2986 printf("\t-l|--label LABEL set filesystem label\n");
2987 printf("\t-L|--copy-label use label from converted filesystem\n");
2988 printf("\t-p|--progress show converting progress (default)\n");
2989 printf("\t-O|--features LIST comma separated list of filesystem features\n");
2990 printf("\t--no-progress show only overview, not the detailed progress\n");
2992 printf("Suported filesystems:\n");
2993 printf("\text2/3/4: %s\n", BTRFSCONVERT_EXT2 ? "yes" : "no");
2996 int main(int argc, char *argv[])
3002 u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
3003 BTRFS_MKFS_DEFAULT_NODE_SIZE);
3006 int usage_error = 0;
3009 char fslabel[BTRFS_LABEL_SIZE];
3010 u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
3013 enum { GETOPT_VAL_NO_PROGRESS = 256 };
3014 static const struct option long_options[] = {
3015 { "no-progress", no_argument, NULL,
3016 GETOPT_VAL_NO_PROGRESS },
3017 { "no-datasum", no_argument, NULL, 'd' },
3018 { "no-inline", no_argument, NULL, 'n' },
3019 { "no-xattr", no_argument, NULL, 'i' },
3020 { "rollback", no_argument, NULL, 'r' },
3021 { "features", required_argument, NULL, 'O' },
3022 { "progress", no_argument, NULL, 'p' },
3023 { "label", required_argument, NULL, 'l' },
3024 { "copy-label", no_argument, NULL, 'L' },
3025 { "nodesize", required_argument, NULL, 'N' },
3026 { "help", no_argument, NULL, GETOPT_VAL_HELP},
3027 { NULL, 0, NULL, 0 }
3029 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
3044 nodesize = parse_size(optarg);
3051 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
3053 "WARNING: label too long, trimmed to %d bytes\n",
3054 BTRFS_LABEL_SIZE - 1);
3056 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
3065 char *orig = strdup(optarg);
3068 tmp = btrfs_parse_fs_features(tmp, &features);
3071 "Unrecognized filesystem feature '%s'\n",
3077 if (features & BTRFS_FEATURE_LIST_ALL) {
3078 btrfs_list_all_fs_features(
3079 ~BTRFS_CONVERT_ALLOWED_FEATURES);
3082 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
3085 btrfs_parse_features_to_string(buf,
3086 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
3088 "ERROR: features not allowed for convert: %s\n",
3095 case GETOPT_VAL_NO_PROGRESS:
3098 case GETOPT_VAL_HELP:
3101 return c != GETOPT_VAL_HELP;
3105 if (check_argc_exact(argc - optind, 1)) {
3110 if (rollback && (!datacsum || noxattr || !packing)) {
3112 "Usage error: -d, -i, -n options do not apply to rollback\n");
3121 file = argv[optind];
3122 ret = check_mounted(file);
3124 fprintf(stderr, "Could not check mount status: %s\n",
3128 fprintf(stderr, "%s is mounted\n", file);
3133 ret = do_rollback(file);
3135 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3136 copylabel, fslabel, progress, features);